senior-devops

Imported from borghei/claude-skills (engineering/senior-devops/SKILL.md). Install upstream with npx skills add borghei/claude-skills --skill senior-devops. Copyright stays with the author (MIT + Commons Clause).

Senior DevOps Engineer

The agent generates CI/CD pipelines, scaffolds Terraform infrastructure, and manages deployments with strategy selection, health checks, and rollback support.

Core Capabilities

• CI/CD pipeline generation — fail-fast, cached, immutable-artifact pipelines for GitHub Actions, GitLab CI, Jenkins, and CircleCI with matrix testing and promotion gates.
• Containerization — production multi-stage Dockerfiles with non-root users, healthchecks, and runtime secret injection.
• Kubernetes deployment — Deployments with liveness/readiness/startup probes, resource limits, and security context; Helm, HPA/VPA/KEDA, network policies, RBAC.
• Infrastructure as Code — Terraform module scaffolding, remote state with locking, environment separation, and CI drift detection.
• Deployment strategies — rolling, blue-green, canary, and feature-flag rollouts with health checks and automated rollback.
• Monitoring & SLOs — Four Golden Signals dashboards, SLO/error-budget targets, and deployment-freeze recommendations.

When to Use

• Building or optimizing a CI/CD pipeline.
• Containerizing an app or deploying to Kubernetes.
• Provisioning cloud infrastructure with Terraform.
• Choosing and executing a deployment strategy (blue-green/canary).
• Handling an infrastructure incident or rollback.

Clarify First

Before generating pipelines or infra, settle the inputs below. When one is missing or vague and would change the result, ask about it; when it wouldn't, use a sensible default and say which one you used.

• CI/CD platform: GitHub Actions / GitLab CI / Jenkins / CircleCI (changes the generated pipeline config).
• Deployment strategy: rolling / blue-green / canary / feature-flag (sets health checks and rollback in the deployment plan).
• Target cloud & IaC scope: provider and which Terraform modules are needed (drives terraform_scaffolder).

Ask about the two or three inputs that matter most for this request, not the whole list. If the user wants a result straight away, produce it and state your assumptions at the top.

Tools

All three are Python standard library only, read local files, make no network calls and never run other programs. Each prints readable text by default and JSON with --json, and exits 1 with a one-line error: message on bad input.

Flags.

• pipeline_generator.py: --platform github-actions|gitlab-ci|circleci|jenkins (default github-actions), --no-docker, --output/-o PATH (relative to the project; refuses to overwrite without --force), --verbose/-v. Without --output it only prints. Version files are honoured only for Node 18 to 26 and Python 3.8 to 3.14 (otherwise Node 22 or Python 3.12, with a warning). GitHub Actions are pinned to major-version tags by design; pin to commit SHAs for high-assurance repositories (see references/cicd_pipeline_guide.md).
• terraform_scaffolder.py: --provider aws|gcp|azure (S3 + DynamoDB, GCS or azurerm backend), --name (default app), --envs (default dev,staging,prod), --force, --dry-run, --verbose/-v. Refuses to write if any target .tf or .tfvars file exists, unless --force; an existing README.md or .gitignore is always kept, never overwritten. Names must be lowercase with single inner hyphens (orders-api). Azure scaffolds include the subscription_id that azurerm 4.x requires, keep the environment suffix in the 24-character storage account name, and limit environments to 12 characters without hyphens. The generated README notes that dev sets force_destroy = true and that bucket names must be globally unique.
• deployment_manager.py: --strategy rolling|blue-green|canary to override the choice, --example to print a sample spec, --output/-o PATH to save the JSON plan, --verbose/-v to include the resolved spec. It only plans; it executes nothing. service must be a Kubernetes name and version an image tag, because both appear in the runbook commands.

Sample inputs in assets/: sample-node-app/ and sample-python-app/ (tiny projects), and deploy-canary.json, deploy-blue-green.json, deploy-rolling.json (one spec per strategy).

Example output.

$ python scripts/pipeline_generator.py assets/sample-node-app
Detected: node 22 (npm), Docker: yes
Stages: install, lint, test, build, docker

# .github/workflows/ci.yml
name: CI
on:
  push:
    branches:
      - main
  pull_request: {}
permissions:
  contents: read
...

$ python scripts/terraform_scaffolder.py ./infra --provider aws --name orders-api --envs dev,prod
Created 18 files (aws, s3 backend, envs: dev, prod)
infra/
  .gitignore
  README.md
  environments/
    dev/
      backend.tf
      main.tf
...

$ python scripts/deployment_manager.py assets/deploy-canary.json
Deployment plan: checkout-api 2.3.1 -> 2.4.0  [canary]
  Why: User-facing, medium risk, 400 rps: enough traffic to judge a small slice before exposing everyone.

Steps:
  1. deploy canary: Start checkout-api:2.4.0 next to the stable version with 0% traffic; wait for readiness.
  2. shift 5% [traffic 5%] [bake 10 min]: Route 5% to 2.4.0; compare canary against stable on the rollback triggers for 10 min.
...
Automatic rollback when any of these holds:
  - error rate (5xx / total) > 0.8% (2 consecutive 1-min windows)
  - p99 latency > 330 ms (3 consecutive 1-min windows)
...

deploy-blue-green.json picks blue-green (high risk, 1-minute rollback tolerance, capacity can double) and deploy-rolling.json picks rolling (low risk, internal worker).

References

Load the reference that matches the task — keep this file lean and pull detail on demand:

• references/devops-workflows-and-operations.md — the three end-to-end workflows (containerize & deploy, Terraform IaC, CI/CD design) with worked Dockerfile/YAML/HCL examples, the deployment-strategy selection matrix and canary ladder, monitoring essentials (Four Golden Signals, SLO targets), anti-patterns, and the troubleshooting table. Read when executing any workflow or diagnosing an incident.
• references/cicd_pipeline_guide.md — stage order, lockfile installs and cache keys per stack, platform differences, least-privilege and pinning defaults, review checklist. Read with pipeline_generator.py output.
• references/infrastructure_as_code.md — module and environment layout, version pinning, remote state and locking per backend, variable validation, plan/apply workflow and drift detection. Read with terraform_scaffolder.py output.
• references/deployment_strategies.md — the strategy selection rules, canary ladder and bake-time formula, probe settings, rollback triggers, runbooks and expand/contract migrations. Read with deployment_manager.py output.
• references/kubernetes_patterns.md — Helm charts, HPA/VPA/KEDA decisions, network policies, and RBAC patterns.
• references/cloud_platform_guide.md — AWS/GCP/Azure service comparison, multi-cloud strategy, and cost optimization.

Scope & Limitations

In Scope:

• Generating a starter CI pipeline from a project's files, scaffolding a Terraform module with environments and remote state, and planning a rolling, blue-green or canary release from a spec (the three scripts).
• Guidance and worked examples for containerizing and deploying an application, Terraform infrastructure as code and CI/CD pipeline design, in references/devops-workflows-and-operations.md.
• Deployment strategy selection (rolling, blue-green, canary, feature flags), monitoring essentials (Four Golden Signals, SLO targets), anti-patterns and troubleshooting.
• Kubernetes patterns (Helm structure, HPA, VPA and KEDA, network policies, RBAC, container security) in references/kubernetes_patterns.md.
• Cloud service comparison, cost optimization, alerting rules and incident runbook templates in references/cloud_platform_guide.md.

Out of Scope:

• Linting existing pipelines, cache tuning and deploy-stage generation (see engineering/ci-cd-pipeline-builder/).
• Static review of Terraform modules and security rules (see engineering/terraform-patterns/) and of Helm charts (see engineering/helm-chart-builder/).
• Running an incident (see engineering/incident-commander/) and designing observability (see engineering/observability-designer/).
• Applying anything to a cloud account or cluster. The skill has no cloud, cluster or CI access.

Important Caveats:

• pipeline_generator.py reads only the project root and generates for the first stack it finds; a monorepo needs one run per package. It stops at the Docker build and does not push images or deploy.
• terraform_scaffolder.py writes placeholder state bucket names (CHANGE-ME) and one starter storage resource; replace both before terraform init.
• deployment_manager.py plans from the numbers in the spec. Rollback thresholds are only as good as the baseline and slo supplied.
• Cloud services, CI platform syntax and Kubernetes APIs change often, so examples need checking against current provider documentation before use.

Integration Points

Last Updated: June 2026 Version: 2.2.0